data_text_search / search_funcs /semantic_ingest_functions.py
seanpedrickcase's picture
Changed embedding model to MiniLM-L6 as faster. Compressed embeddings are now int8. General improvements to API mode
ea0dd40
raw
history blame
7.06 kB
import time
import ast
import gzip
import pandas as pd
import gradio as gr
import pickle
from typing import Type, List, Literal
from pydantic import BaseModel, Field
# Creating an alias for pandas DataFrame using Type
PandasDataFrame = Type[pd.DataFrame]
PandasSeries = Type[pd.Series]
class Document(BaseModel):
"""Class for storing a piece of text and associated metadata. Implementation adapted from Langchain code: https://github.com/langchain-ai/langchain/blob/master/libs/core/langchain_core/documents/base.py"""
page_content: str
"""String text."""
metadata: dict = Field(default_factory=dict)
"""Arbitrary metadata about the page content (e.g., source, relationships to other
documents, etc.).
"""
type: Literal["Document"] = "Document"
from search_funcs.helper_functions import get_file_path_end, ensure_output_folder_exists
from search_funcs.bm25_functions import save_prepared_bm25_data, output_folder
from search_funcs.clean_funcs import initial_clean
def combine_metadata_columns(df:PandasDataFrame, cols:List[str]) -> PandasSeries:
'''
Construct a metadata column as a string version of a dictionary for later parsing.
Parameters:
- df (PandasDataFrame): Data frame of search data.
- cols (List[str]): List of column names that will be included in the output metadata column.
Returns:
- PandasSeries: A series containing the metadata elements combined into a dictionary format as a string.
'''
df['metadata'] = '{'
df['blank_column'] = ''
for n, col in enumerate(cols):
df[col] = df[col].astype(str).str.replace('"',"'").str.replace('\n', ' ').str.replace('\r', ' ').str.replace('\r\n', ' ').str.cat(df['blank_column'].astype(str), sep="")
df['metadata'] = df['metadata'] + '"' + cols[n] + '": "' + df[col] + '", '
df['metadata'] = (df['metadata'] + "}").str.replace(', }', '}').str.replace('", }"', '}')
return df['metadata']
def clean_line_breaks(text:str):
'''Replace \n and \r\n with a space'''
return text.replace('\n', ' ').replace('\r', ' ').replace('\r\n', ' ')
def parse_metadata(row):
'''
Parse a string instance of a dictionary into a Python object.
'''
try:
# Ensure the 'title' field is a string and clean line breaks
#if 'TITLE' in row:
# row['TITLE'] = clean_line_breaks(row['TITLE'])
# Convert the row to a string if it's not already
row_str = str(row) if not isinstance(row, str) else row
row_str.replace('\n', ' ').replace('\r', ' ').replace('\r\n', ' ')
# Parse the string
metadata = ast.literal_eval(row_str)
# Process metadata
return metadata
except SyntaxError as e:
print(f"Failed to parse metadata: {row_str}")
print(f"Error: {e}")
# Handle the error or log it
return None # or some default value
def csv_excel_text_to_docs(df:PandasDataFrame, in_file:List[str], text_column:str, clean:str = "No", return_intermediate_files:str = "No", progress=gr.Progress(track_tqdm=True)) -> tuple:
"""Converts a DataFrame's content to a list of dictionaries in the 'Document' format, containing page_content and associated metadata.
Parameters:
- df (PandasDataFrame): Data frame of search data.
- in_file (List[str]): List of input file names.
- text_column (str): The text column that will be searched.
- clean (str): Whether the text is cleaned before searching.
- return_intermediate_files (str): Whether intermediate processing files are saved to file.
- progress (gr.Progress, optional): The progress tracker for the operation.
Returns:
- tuple: A tuple containing data outputs in a Document class format, an output message, and a list of output file paths.
"""
ensure_output_folder_exists(output_folder)
output_list = []
if not in_file:
return None, "Please load in at least one file.", output_list
progress(0, desc = "Loading in data")
file_list = [string.name for string in in_file]
data_file_names = [string for string in file_list if "tokenised" not in string and "npz" not in string.lower()]
if not data_file_names:
return doc_sections, "Please load in at least one csv/Excel/parquet data file.", output_list
if not text_column:
return None, "Please enter a column name to search", output_list
data_file_name = data_file_names[0]
# Check if file is a document format, and explode out as needed
if "prepared_docs" in data_file_name:
print("Loading in documents from file.")
doc_sections = df
# Convert each element in the Series to a Document instance
return doc_sections, "Finished preparing documents", output_list
ingest_tic = time.perf_counter()
doc_sections = []
df[text_column] = df[text_column].astype(str).str.strip() # Ensure column is a string column
original_text_column = text_column
if clean == "Yes":
progress(0.1, desc = "Cleaning data")
clean_tic = time.perf_counter()
print("Starting data clean.")
df_list = list(df[text_column])
df_list = initial_clean(df_list)
# Save to file if you have cleaned the data. Text column has now been renamed with '_cleaned' at the send
out_file_name, text_column, df = save_prepared_bm25_data(data_file_name, df_list, df, text_column)
df[text_column] = df_list
clean_toc = time.perf_counter()
clean_time_out = f"Cleaning the text took {clean_toc - clean_tic:0.1f} seconds."
print(clean_time_out)
cols = [col for col in df.columns if col != original_text_column]
df["metadata"] = combine_metadata_columns(df, cols)
progress(0.3, desc = "Converting data to document format")
# Create a list of Document objects
doc_sections = [Document(page_content=row[text_column],
metadata= parse_metadata(row["metadata"]))
for index, row in progress.tqdm(df.iterrows(), desc = "Splitting up text", unit = "rows")]
ingest_toc = time.perf_counter()
ingest_time_out = f"Preparing documents took {ingest_toc - ingest_tic:0.1f} seconds"
print(ingest_time_out)
if return_intermediate_files == "Yes":
progress(0.5, desc = "Saving prepared documents")
data_file_out_name_no_ext = get_file_path_end(data_file_name)
file_name = data_file_out_name_no_ext
if clean == "No":
out_doc_file_name = output_folder + file_name + "_prepared_docs.pkl.gz"
with gzip.open(out_doc_file_name, 'wb') as file:
pickle.dump(doc_sections, file)
elif clean == "Yes":
out_doc_file_name = output_folder + file_name + "_cleaned_prepared_docs.pkl.gz"
with gzip.open(out_doc_file_name, 'wb') as file:
pickle.dump(doc_sections, file)
output_list.append(out_doc_file_name)
print("Documents saved to file.")
return doc_sections, "Finished preparing documents.", output_list